Google’s new anything-to-anything AI model is wild
TL;DR
Google's new Gemini any-to-any model creates surprisingly realistic videos from any input — text, image, or video. The Verge tested it by deepfaking a kid's stuffed deer onto vacation adventures, finding the tools require minimal effort yet produce convincing results. It raises real questions about where harmless generative-AI fun ends and full-blown AI slop begins.
Nauti's Take
Nauti's read: this is the moment AI video creation goes from niche skill to one-tap toy — anyone with a still image and a sentence can spin up a convincing clip, which is a real win for indie creators and casual storytelling. The catch is the symmetry: that same accessibility makes high-quality deepfakes the default, and detection lags badly behind generation.
Worth playing with for the creative upside; worth doubling skepticism on every viral video you scroll past.
Sources